On 28/06/13 14:42, Sergey Kirpichev wrote:
> There is a long term issue [1] to make mpmath an external dependency. 
> It was suggested by
> Aaron (see discussion in PR [2]), that we should discuss this on the
> mailing list.
> 
> Perhaps, the main reason to prefer bundled mpmath copy is a
> mystical "complex installation".  Not clear enough.  For example, I don't
> think that additional steps for package installation from mpmath manual
> does matter.  For non-root users
> there shouldn't be any problems too, at least for Python 2.6+ (per-user
> site-package directory).
> 
> [1] http://code.google.com/p/sympy/issues/detail?id=2482
> [2] https://github.com/sympy/sympy/pull/2192

As someone who has an application which uses *both* SymPy *and* mpmath I
am absolutely against removing the embedded copy.  There are several
reasons for this.

When we first started developing our application the plan was to allow
for both system and embedded mpmath installations.  The result was code
along the lines of:

try:
    import mpmath as mp
except ImportError:
    from sympy import mpmath as mp

which was something of a nuisance.  Worse, if the user had opted to
install their own version of mpmath but SymPy was configured to use its
own version things were found to go seriously awry.  Eventually we gave
up and just switched to using the embedded copy within SymPy.  Removing
mpmath would reopen this can of worms for very little benefit.

Now, if mpmath was a flourishing project I would be open to the idea of
splitting the two apart.  However, as of June 2013, it isn't.  There
have been no commits to the official subversion repository in quite some
time (Nov 2011) with the version that ships with SymPy, for all intents
and purposes, being the de facto revision.  Case in point: Google
mpmath, the second result for me is:

http://docs.sympy.org/dev/modules/mpmath/index.html

which is from the SymPy project.

I would much rather see mpmath brought back under the umbrella of SymPy
at least until a new maintainer can be found.  At the very least I do
not believe that removing the embedded copy of mpmath should even be on
the table until all of the patches (or functional equivalents thereof)
have been merged upstream and a new mpmath release has been made.

Polemically yours, Freddie.

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to